Skip to content

agency-agents: role scaffolds for subagents

AI tools · Subagents
Not studied

Added 2026-10-29.

Problem

A subagent needs a role: an area of responsibility, a way of working, criteria for the result. Writing each one from scratch takes a long time.

Solution

The project calls itself "a complete AI agency at your fingertips", where "each agent is a specialized expert with personality, processes, and proven deliverables". It has hundreds of markdown roles with a uniform frontmatter, including DevOps Automator, SRE, Platform Engineer and a Security division. Take one as a base and rewrite it for your stack. The project's CI checks the structure and rejects recolored copies of existing roles.

Example

git clone https://github.com/msitarzewski/agency-agents.git
cp agency-agents/engineering/engineering-sre.md ~/.claude/agents/

Then, in the copied role:

  • replace the generic tool list with your stack: GitLab CE, Docker, Ansible, Prometheus;
  • add restrictions: what the agent is not allowed to do in production;
  • write down the done criteria for the result.

Limitations

These are prompt personas, not proven practices: the "proven deliverables" and production-readiness claims aren't backed by anything, and CI checks the form, not the behavior. The repository's popularity doesn't reflect its usefulness. Run agent-scan before installing.

Source

How to apply it yourself

  1. Take the SRE and DevOps Automator roles.
  2. Rewrite them for your stack and restrictions.
  3. Test them on three real tasks and compare with working without the role.
  4. Mark it "Studied" if the role noticeably improved the result.